PARENT SESSION
Poster Session #60: Statistical Ecology.
Thursday, August 8. Presentation from 8:00 AM to 9:30 AM. Exhibit Hall B & C, TCC

Making the most of a sparse data situation: Model selection and parameter estimation from count data.

Omland, Kristian^*,1, ¹ Union College, Schenectady, NY, 12308

ABSTRACT- Census data from populations of imperiled species and pests alike offer a rich opportunity for inference about processes to include in a forecasting model and parameter estimates for the model. I used maximum likelihood to fit a set of candidate models to monitoring data for the Puritan tiger beetle, and imperiled species on the Connecticut River. Data were counts (relative abundance) from two metapopulations over 12 and 16 years. I fit the models to the data assuming alternatively a deterministic process imperfectly observed (observation uncertainty) or a stochastic process perfectly observed (process uncertainty). Information theoretic model selection criteria and likelihood ratio tests revealed support for an exponential model with no movement among patches, a model that included isotropic movement, or a model that included a population ceiling. Model averaging permitted accounting for model selection uncertainty (in addition to observation and process uncertainty) when estimating parameters of the model. Estimated rate of population change was positive regardless of the model, although 95% confidence intervals did not exclude negative change. Allowing for movement led to a lower estimated rate of population change, while allowing for density-dependent growth led to a higher estimate. I incorporated fragmentary demographic data into the model fitting by 1) constraining the maximum likelihood fitting and 2) adopting a Bayesian approach with Markov chain Monte Carlo analysis. Neither significantly improved parameter estimation in terms of reduced uncertainty.

KEY WORDS: time series, population models, model selection, parameter estimation